test_gdb's get_gdb_repr carves up a gdb backtrace to try to extract how gdb representated the data.

When connected to a tty, gdb will insert additional newlines and spaces based on the width of the tty (internally it has a wrap_here() function to do this), so the test turned out to be somewhat susceptible to whitespace and tty configuration.

I'm attaching a patch against trunk which I believe fixes this.  I've tested it with various tty widths (from 5 columns wide through to 235 columns wide), and redirecting to a file, and all tests pass.

[Seen on buildbot on this run:
http://www.python.org/dev/buildbot/builders/sparc%20Ubuntu%20trunk/builds/37/steps/test/logs/stdio
and I believe this was the cause of all of four of the five failures there.

The remaining one "test_corrupt_tp_name" seems to be a different issue]